Sketches 2x2 taxonomy of alignment targets sorting models by controllability and value alignment.
How media typically covers Theia Vogel
Theia Vogel as author
Large language models consistently hallucinate a non-existent seahorse emoji and exhibit erratic behavior when attempting to produce it, revealing interpretability insights about how LLMs handle false latent beliefs.
“Author of "Why Do LLMs Freak Out Over the Seahorse Emoji?"”
Referenced in coverage
“Sketches 2x2 taxonomy of alignment targets sorting models by controllability and value alignment.”